The effect of the variability of wind forcing on ENSO simulation in an OGCM: case of canonical and protracted event

This study is an attempt to understand the onset and the evolution of canonical (typical length ~ 18–24 months; CE) and protracted El Niño (> greater than 3 years; PE) compared to the normal state (NS) in an ocean model. Indo-Pacific warm pool indicates higher sea surface temperature (SST) before the onset of strong CE compared to the NS and PE. The ocean model (MOM5.1.0) used in the study shows a systematic SST bias in the Indo-Pacific Ocean with warmer (cooler) temperatures in the western (eastern) Pacific during NS, CE, and PE exhibiting La Niña–like conditions. The model also exhibits deeper thermocline depth in the western equatorial Pacific Ocean during PE and CE than NS, indicating higher heat content values. Despite higher heat content in the western Pacific before the onset of El Niño, the difference in the variability of surface wind forcing during the preceding months determines the type of El Niño. The difference in surface wind forcing among the NS, PE, and CE states without altering the ocean state can modify the subsurface propagation in the equatorial Pacific Ocean. A change in longitudinal extent of upwelling Kelvin waves from western Pacific towards eastern Pacific along with the change in surface wind forcing decides the fate of El Niño. Based on the results of model experiments for 1948–2009, observed features of the recent protracted El Niño of 2014–2016 appear to be a blend of PE and CE in terms of ocean dynamics and surface wind forcing.


Introduction
At the beginning of the twenty-first century, the world experienced an "extended" El Niño during 2014-2016, which falls under the category of protracted El Niño events (Allan et al. 2020). A typical (canonical) El Niño event lasts for ~ 18-24 months. It makes a transition to La Niña or a neutral state within a year after the peak phase of the event, occurring mostly in the boreal winter. Whereas prolonged or protracted El Niño persisting for more than 3 years is not followed by a neutral or cold phase event (Rasmusson and Carpenter 1983;Allan and D'Arrigo 1999), strong easterly wind bursts during June-July of 2014 led to the weakening of the eastern equatorial Pacific warming, and the buildup of equatorial heat content in 2014 provided favorable background conditions for the formation of strong El Niño in 2015 (Levine and McPhaden 2016;Lim et al. 2017). The uncoupled El Niño of 2014 showed a weaker zonal gradient of SST anomaly across the tropical Pacific than canonical El Niño (Yang et al. 2018;Hu et al. 2020). Any change in the tropical SST anomalies can strongly impact El Niño Southern Oscillation (ENSO) teleconnections (Capotondi et al. 2015;Zhao et al. 2021). While the 2015-2016 event reported earlier occurrence of equatorial Pacific warming, the 1997-1998 event showed rapid growth due to stronger westerly wind bursts (WWBs) and the Madden-Julian oscillation during the spring season (Lim et al. 2017). Kakatkar et al. (2018), using reanalysis data and model experiments, compared two strong El Niños of 1997 and 2015 and found that not the strength of El Niño, but large-scale wind forcing and the associated ocean dynamics govern the strength of La Niña in the following year. The equatorial Pacific Ocean witnessed peculiar ocean dynamics, zonal phase propagation, WWBs, and discharge-recharge during this extreme El Niño compared to previous El Niño events (Gasparin and Roemmich 2016;Levine and McPhaden 2016). The ENSO events exhibit diverse behavior in terms of mean state and interannual variability. The variable relative contribution of physical processes underlying the different El Niño can lead to different SST anomalies at different longitudes in the equatorial Pacific (Capotondi et al. 2015;Chen et al. 2016).
Wind stress forcing plays a key role in determining the size and timing of these ENSO events (Bjerknes 1969). WWBs can advect the western Pacific warm pool with SST greater than 28 °C into the central Pacific via the generation of downwelling Kelvin waves along the thermocline across the equatorial Pacific Ocean (Gadgil et al. 1984;McPhaden 1999). Multiple WWBs that occur during the boreal summer and fall months throughout the growth phase of the El Niño can trigger strong El Niño events (McPhaden 1999;Eisenman et al. 2005). Another necessary but not sufficient condition for the onset of ENSO is a build-up of anomalous warm water volume (WWV) over the upper part of the tropical Pacific Ocean (Wyrtki 1985;Jin 1997;Meinen and McPhaden 2000). This build-up of WWV, along with the presence of WWBs during boreal spring and summer monsoon months over the western Pacific, can lead to an El Niño event in the following winter months (McPhaden 2003;Yu and Kao 2007). Both WWV and WWB anomalies over the upper layers of the tropical Pacific Ocean are critical for predicting extreme El Niño events (1982, 1997, and 2015) (Levine and McPhaden 2016). Also, protracted ENSO events (1990 and show unusually large WWV anomalies during the boreal spring season in the absence of significant wind stress forcing during the preceding boreal summer and fall months (Allan and D'Arrigo 1999;Levine and McPhaden 2016;Arora and Kumar 2019). The effect of wind stress forcing on WWV is a function of the current state of the ocean-atmosphere system and the time of the year. WWBs occurring west of the dateline in the boreal winter months led to a build-up of WWV in 1990 and 2014 (Fedorov et al. 2015;Levine and McPhaden 2016;Lim et al. 2017). Arora and Kumar (2019), using observations and reanalysis data for 1980-2010, attempted to understand the process of evolution and decay of the two types (i.e., canonical and protracted) of El Niño. This study hypothesized that frequently (occasionally) occurring weak (strong) WWBs over the western Pacific and Maritime continent before the onset of El Niño can result in a protracted (canonical) El Niño. This study assumed the heat content over the western Pacific to be the same before the onset of both types of El Niño. The anomalous easterlies in the western Pacific during the decay phase of canonical El Niño cool the SST in the eastern Pacific via the generation of upwelling Kelvin waves (Arora et al. 2016). However, weak localized warming in the Indian Ocean is incapable of generating a significant atmospheric response in the form of easterlies (Arora and Kumar 2019). Though many climate models capture the ENSO variability reasonably well, there are still biases in simulation and prediction of ENSO (Ren et al. 2016;Zheng et al. 2016;Arora et al. 2018). Earlier studies have emphasized the forcing effects of wind stress, heat flux, and freshwater flux forcing on ENSO modulations in the coupled system (Waliser et al. 1994;Xie and Philander 1994;Gasparin and Roemmich 2016). This study focuses on the role of horizontal wind forcing variability at the ocean surface without altering the mean state in an ocean general circulation model during normal, protracted, and canonical ENSO to test the hypothesis proposed in Arora and Kumar (2019). The paper is organized as follows: Section 2 describes the model experiment used in the study. Section 3 discusses the study results, and Section 4 summarizes the study's major findings. The criterion of choosing these different ocean states is explained in Section 3.

The model used and details of the experiment
The ocean general circulation model (OGCM) used in this study is the Geophysical Fluid Dynamics Laboratory (GFDL) Modular Ocean Model version 5.1.0 (MOM5.1.0). This model provides a numerical representation of the primitive equations used for studying the ocean climate system (Griffies et al. 2004). This model is a global model with 1° zonal resolution and the meridional resolution varies from 0.33° at the equator to 0.67° in the tropics and gradually increasing to 1° towards the pole. There are 50 vertical levels from the surface to 5000m depth with a 10-m resolution in the first 250m. The pressure-based vertical coordinate system is used in the model. MOM5.1.0 includes the Arakawa C-grid option as well as a dynamically interacting Lagrangian sub-model. MOM5.1.0 is initialized using the annual climatology of temperature and salinity from Levitus (1998) and is forced with the GFDL Coordinated Ocean-ice Reference Experiments (CORE-V2) normal year forcing of downwelling shortwave and longwave radiation, surface wind fields at 10m, specific humidity, air temperature, surface pressure, and surface precipitation (Large and Yeager 2009) and is spun up for 100 years. The CORE-V2 forcing is available at https:// data1. gfdl. noaa. gov/ nomads/ forms/ core/ COREv2. html. From this state, the model is forced with CORE-V2 interannually varying forcing for 1948-2009. This is termed as control run (CTRL) in the study. For wind sensitivity experiments with OGCM, the model state in the CTRL at the end of the year 1957 (pre-NS), the year 1979 (pre-CE), the year 1989 (pre-PE), and the year 1995 (pre-CE1) are used as reference states for experimental runs. The reason for referencing these years as pre-NS, pre-CE, pre-PE, and pre-CE1 is explained in Section 3.1. The model is integrated for 5 years from 1990 to 1994 from the pre-PE state using wind forcing used during the canonical period (CE) and normal period (NS) in the control run. This is termed as exp1 and exp2, respectively, in the study. Similar experiments are performed by interchanging the surface wind forcing during the CE (NS) period with the surface wind forcing used during the CTRL in PE and NS (CE). Details of all OGCM experiments are provided in Table 1. The wind forcing at 10m is extended until 2020 using NCEP-DOE reanalysis-II available at https:// psl. noaa. gov/ data/ gridd ed/ data. ncep. reana lysis2. gauss ian. html to compare the results of model experiments with the recent protracted El Niño of 2014-2016 (Kanamitsu et al. 2002). The model results are compared further with instrument subsurface ocean temperature data EN4-v4.2.1 from Met Office Hadley Centre available at https:// www. metoffi ce. gov. uk/ hadobs/ en4/ for 1948-2020 (Good et al. 2013).

Mean ocean state during different types of El Niño
A warming event in the Pacific Ocean is termed as El Niño when the 3-month running mean of the SST anomaly averaged over the Niño3.4 region (170° W-120° W, 5° S-5° N) crosses a threshold of 0.5 °C, for at least five consecutive seasons. Climate Prediction Center-National Oceanic and Atmospheric Administration (CPC-NOAA) uses the same definition for their operational forecasts. An El Niño can be categorized into canonical and protracted ENSO based on the persistence of area-averaged SST anomaly over the Niño3.4 region. If a positive SST anomaly during an El Niño event lasts for more (less) than 3 consecutive years, the event is termed as protracted (canonical) El Niño (Allan and D'Arrigo 1999). Figure 1 shows the seasonal mean (3-month running mean) of SST anomaly averaged over the Niño3.4 region for 1948-2020.  Allan et al. (2020). These prolonged warm states of the equatorial ocean are termed protracted El Niño states (1990PE and2013-2017;PE1). Since warming associated with PE events typically lasts for 3 years, 5 years is considered (starting T i − 1 until T e + 1, where T i and T e are the years of onset and withdrawal of any type of El Niño event, respectively) to capture the growth and decay of El Niño event in the equatorial Pacific. Observed SST anomaly averaged over the Niño3.4 region shows the transition of strong canonical El Niño in 1982-1983to a La Niña in 1983-1984and 1984-1985 For robustness of the findings, the analysis results of the canonical El Niño evolution of 1982-1983 are also compared with another typical strong canonical El Niño event of 1997-1998, followed by La Niña in 1998-1999. In this study, this state is termed as canonical El Niño state (1980-1984CE and1996-2000;CE1). Peak SST anomalies during PE1 (~ 2.5 °C) are significantly higher than PE (1.6 °C) and are comparable to CE1 (~ 2.4 °C). These values are in close agreement with Puy et al. (2017). The ENSO of 2015-2016 is considered one of the three strongest events (1982-1983, 1997-1998, and 2015-2016) by many previous studies (Bell et al. 2016;Huang et al. 2016). SST anomalies averaged over the Niño3.4 region during 1958-1962 remained less than the threshold value (± 0.5 °C) of defining an ENSO event except for a very weak borderline El Niño event during 1958-1959 with peak SST anomaly reaching 0.6 °C. There was no strong El Niño or La Niña in the eastern Pacific during 1958-1962. Therefore, this state is termed the normal state (1958-1962NS). The chosen ocean states in the study match well with the ENSO information from https:// origin. cpc. ncep. noaa. gov/ produ cts/ analy sis_ monit oring/ ensos tuff/ ONI_ v5. php and Li et al. (2021). The ocean state before the start of NS (December 1957), CE (December 1979), and PE (December 1989), and CE1 (December  1990-1994 1980-1984 exp2 1990-1994 1958-1962 exp3 1980-1984 1990-1994 exp4 1980-1984 1958-1962 exp5 1958-1962 1990-1994 exp6 1958-1962 1980-1984 exp7 1996-2000 1958-1962 exp8 1996-2000 1990-1994 Fig. 1 Seasonal running mean (3 months) of SST anomalies averaged over the Niño3.4 region for 1948-2020. The solid black line at ± 0.5 °C represents the threshold value for El Niño or La Niña 1995) is termed as pre-NS, pre-CE, pre-PE, and pre-CE1, respectively. Figure 2(a-d) shows mean SST during pre-NS, pre-CE, pre-PE, and pre-CE1 state in CTRL, respectively. The difference in the spatial structure of mean SST in the tropical Indo-Pacific region indicates the different heat content values before the onset of different ENSO types. Each state exhibits a peculiar signature of zonal extent and position of the Indo-Pacific warm pool (Gadgil et al. 1984;Graham and Barnett 1987). Oceanic anomalies before the onset of El Niño over the western Pacific are related to the strengthening of the trades, and the accumulated warm water flows eastward in the form of Kelvin waves to initiate an El Niño (Wyrtki 1975). Also, the cold tongue in the equatorial eastern Pacific extends towards the international dateline in the pre-PE state than the pre-NS and pre-CE. Figure 2(e and f) shows the difference in the mean SST of pre-CE and pre-PE state from the pre-NS state as shown in Fig. 2(a-c). Higher values of SST are seen in the Indo-Pacific warm pool during the pre-CE and pre-PE state before the onset of strong canonical ENSO compared to the pre-NS state. The Bay of Bengal region shows cooler SST during pre-PE state compared to pre-NS and pre-CE state ( Fig. 2(f-g)). A comparison of pre-CE and pre-PE state shows the higher SST (by 2°C) in the cold tongue region during pre-CE conditions with cooler SST over the western Pacific (Fig. 2g). It implies that heating is confined to the western Pacific before the onset of the protracted ENSO with additional cooling along the eastern Pacific compared to canonical ENSO. The difference of mean SST during pre-CE1 and pre-CE SST before the onset of identical canonical ENSO events (CE1 and CE) shows enhanced (reduced) values of SST over the equatorial eastern (central) Pacific Ocean before the start of El Niño (Fig. 2h). This difference in the mean ocean state is consistent with the slightly higher values of SST anomalies during the 1997-1998 event (2.4°C) compared to [1982][1983] (2.2°C), as shown in Fig. 1.
To understand the model's fidelity in simulating the mean state, the mean structure of the tropical SST is seen in the model control run and compared with observations. Figure 3 shows the mean SST during NS, CE, PE, and CE1 events for CTRL ( Fig. 3(a-d)) and observations ( Fig. 3(e-h)) and their . Higher values of SST are seen over the Indo-Pacific warm pool region with the extent of the warm pool from the eastern Indian Ocean to the western Pacific Ocean. A comparison of the mean SST in CTRL and observations during NS shows warmer (cooler) SST in the western (eastern) Pacific Ocean near the dateline in the CTRL (Fig. 3(i)). Also, the mean SST is cooler than the observed SST over the Arabian Sea and the Bay of Bengal region in CTRL. This structure of the model bias is systematic and appears during CE and PE with a small difference in magnitude (Fig. 3(j and k)). This systematic bias exists in CE1 too, but of a relatively low magnitude compared to CE (Fig. 3(l)). This systematic bias in SST in the OGCM leads to an increased east-west gradient of SST along the equator in the Pacific Ocean, resembling spurious La Niña-like conditions compared to observations. Since zonal SST gradients are coupled to zonal atmospheric circulation, known as the Walker circulation, any change in zonal SST gradient can further alter the zonal tilt of the oceanic thermocline and strength of the equatorial upwelling (Allan and D'Arrigo 1999;Levine and McPhaden 2016;Arora and Kumar 2019;Hu et al. 2020;Zhao et al. 2021). Figure 4 shows the difference of SST for the mean states of CE, PE, and NS states for the CTRL and observations, as shown in Fig. 3. The difference of mean SST during CE and NS state shows warmer SST (~2°C) over the eastern Pacific and the Indian Ocean in the CTRL and observations (Fig. 4(a and d)). The difference of mean SST during PE and NS shows wider warmer anomalies in the eastern Pacific (~1.6°C) extending up to the central Pacific in the CTRL and observations ( Fig. 4(b and e)). The OGCM used in the study can capture the typical signature of SST anomalies during canonical and The subsurface ocean state in CTRL and observations is analyzed to get further insight into the ocean dynamics. Figure 5 shows the mean depth of 20°C isotherm (D20) during NS, CE, PE, and CE1 events for CTRL ( Fig. 5(a-d)) and observations (Fig. 5(e-h)) and their respective differences ( Fig. 5(i-l)). OGCM can simulate the subsurface structure of the thermocline very well consistent with the surface. The CTRL can capture the spatial structure of shallow thermocline regions in the Western Indian Ocean (the Seychelles Dome) and eastern Pacific (Yokoi et al. 2008). However, the value of D20 in deeper thermocline regions has a positive bias in the western and southern tropical Indian Ocean and the northern Pacific during all types of ENSO events (NS, CE, and PE) (Fig. 5(i-k)). This positive bias is strongest in NS events (Fig. 5k). Also, D20 is underestimated (overestimated) in PE and CE (NS) events in the southwestern and central Pacific. The systematic subsurface bias seen in CE is also seen in CE1 with similar magnitude, unlike surface bias (Figs. 3 and 5). Figure 6 shows the difference of mean D20 for CE, PE, and NS states for CTRL and observations, as shown in Fig. 5. The thermocline depth is deeper (shallower) in the eastern (western and central) Pacific Ocean during CE state compared to NS in CTRL and observations ( Fig. 6(a and d)). The difference of the mean D20 for PE and NS state shows a signature similar to the difference of the mean D20 for CE and NS state (Fig. 6a) with the extension of shallower D20 restricted to western Pacific in observations (Fig. 6e). This reduced extent of the shallow thermocline restricted to the western Pacific is consistent with the zonal extension of warmer mean SST from the eastern Pacific towards the International dateline during PE compared to CE (Fig. 4f). However, the eastward extension of the deeper (shallower) thermocline is confined to the eastern (western and central) Pacific in PE compared to NS in CTRL (Fig. 6b), consistent with the shallower thermocline bias in central Pacific in PE (Fig. 5k). A deeper thermocline depth in the eastern Pacific during PE and CE compared to NS is a qualitative measure of the higher value of WWV acting as a precursor for ENSO initiation (Arora and Kumar 2019). Also, higher values of D20 seen over the western Pacific during PE compared to CE events (Fig. 6(c and f)) are consistent with previous studies (Levine and McPhaden 2016;Arora and Kumar 2019).

Preconditioning of equatorial Pacific for triggering of ENSO
The coupled phenomenon of ENSO is triggered by the alterations in the zonal SST gradient and associated atmospheric circulation via a change in the tilt of the oceanic thermocline (Bjerknes 1969;Levine and McPhaden 2016;Hu et al. 2020;Zhao et al. 2021). The strength and frequency of WWBs  (Fig. 7a). Zonal wind forcing anomaly at 10m shows predominant westerlies with maximum values (reaching up to 6 ms -1 ) at the end of 1982, yielding the peak of strong El Niño (Fig. 7b). Along with the change in phase of SST anomalies averaged over the Niño3.4 region (Fig. 1), zonal wind anomalies over EWPO also change to easterlies following Bjerknes's (1969) positive feedback in the equatorial Pacific Ocean. A similar evolution of anomalous zonal wind forcing over the EWPO is seen for another canonical (CE1) event. The wind forcing over the EWPO during CE1 shows strengthening of westerlies 1 year prior to the peak of 1997-1998 El Niño (Fig. 7d). During the withdrawal phase, these westerlies reverse sign leading to a strong La Niña in the following year. The difference in the strength of zonal wind anomalies over EWPO during CE and CE1 leads to a difference in SST anomalies over the Niño3.4 region.
The accumulated effect of westerlies over EWPO before the peak of El Niño during CE1 leads to higher values of SST anomaly averaged over the Niño3.4 region, as seen in Fig. 1. This result is consistent with the findings of Arora and Kumar (2019). Wind forcing anomaly during PE event hovers around the climatological mean with values ranging between −4 and 4 ms -1 (Fig. 7c). Similar to comparing the two canonical events, the evolution of wind anomaly averaged over the EWPO during PE is compared with the evolution of observed wind anomalies during another recent protracted event (PE1). Figure 7(e) shows the evolution of zonal wind anomalies averaged over the EWPO during PE1. The positive value (~2 ms -1 ) of zonal wind anomalies averaged over the EWPO at the beginning of 2014 shifts towards the mean value in the following months. The persistent westerlies with a maximum value of 4 ms -1 are seen over the EWPO during 2015. The shorter span of anomalous westerlies followed by anomalous easterlies in 2014 led to a weak El Niño in 2014, while the persistent westerlies led to a strong El Niño in 2015. The difference in anomalous zonal wind evolution patterns during these two protracted El Niño events led to a different evolution of SST anomalies over the Niño3.4 region (Fig. 1).

Fig. 5 Mean D20 for tropical ocean averaged for (a) NS, (b) CE, (c) PE, and (d) CE1 for CTRL. (e-h) same as (a-d) but for observations. i-l shows the difference of D20 for CTRL (a-d) and observations (e-h) respectively
To get further insight into the statistics of WWBs, the histogram of near-surface zonal wind anomalies averaged over the EWPO during different states of the ocean is constructed. Figure 8 shows the histogram of the anomalous zonal wind at 10m averaged over EWPO during NS, CE, PE, CE1, and PE1 states. The frequency of occurrence of anomalous westerlies is substantially lower in NS than PE and CE events ( Fig. 8(a-c)). The histogram of the anomalous zonal wind forcing averaged over the EWPO during NS state shows clear skewness towards the negative values, indicating more anomalous easterlies during NS (Fig. 8a). The histogram of zonal wind anomaly averaged over the EWPO follows a distribution close to normal for PE state (Fig. 8c). However, the distribution is skewed towards positive values during CE and CE1 (Fig. 8b and d). This implies that the difference in zonal wind forcing at the surface in the EWPO before the onset of El Niño can trigger different states of ENSO in the eastern Pacific. However, zonal wind anomalies during PE1 show slight skewness towards positive values, unlike PE, due to the persistent anomalous westerlies over EWPO during 2015 (Fig. 8e), also evident in Fig. 7e.
The build-up and discharge of anomalously warm water over the upper part of the tropical Pacific Ocean, along with the presence of WWBs over the western Pacific, can lead to the onset of an El Niño (Wyrtki 1985;Jin 1997;Meinen and McPhaden 2000;McPhaden 2003;Yu and Kao 2007). Warm water volume (WWV) is defined as the volume of water up to D20 and is an important predictor of ENSO (Meinen and McPhaden 2000;Bunge and Clarke 2014). WWV is calculated for the equatorial eastern Pacific (EP; 5° S-5° N, 80° W-155° W), western Pacific (WP; 5° S-5° N, 120° E-155° W), and entire Pacific Ocean (PO; 5° S-5° N, 120° E-80° W). Figure 9 compares anomalous WWV for the CTRL with observed WWV anomalies calculated using EN4-v4.2.1 averaged over PO, EP, and WP. The CTRL captures the anomalous WWV averaged over the WP region with a high correlation with observations during NS, CE, PE, and CE1 (0.79, 0.8, 0.75, and 0.92, respectively) ( Fig. 9(i-l)). The correlation of WWV anomaly averaged over PO (EP) among observations and CTRL is 0. 32, 0.67, 0.41, and 0.88 (0.3, 0.19, 0.54, and 0.85) during NS, CE, PE, and CE1 respectively. All values except 0.19 (Fig. 9f) are statistically significant at 90%. WWV anomaly averaged over PO and WP remains below normal during NS for CTRL and observations indicating no anomalous build-up of heat content ( Fig. 9a and i). The anomalous WWV averaged over EP hovers around the mean value during NS, indicating no (d-f) same as (a-c) but for observations ENSO-like conditions in EP (Fig. 9e). During the CE event, WWV anomaly averaged over PO and WP increased before the peak of the El Niño event and then decreased as El Niño made the transition to La Niña ( Fig. 9b and j). The anomalous WWV averaged over CE starts increasing from the beginning of 1982. It reaches maximum value around the end of 1982 and then reverses its sign with the transition of El Niño to La Niña. A similar transition of anomalous WWV is seen for the CE1 event for observations and CTRL (Fig. 9d, h, and l). The CTRL exhibits the highest skill in capturing CE1 compared to NS, PE, and CE in all domains of the equatorial Pacific. This enhanced skill can be attributed to the reduced surface temperature bias in CE1 compared to NS, PE, and CE in the CTRL (Fig. 3(i-l)).
Arora and Kumar (2019) highlighted a high correlation value (0.8) between the SST anomaly averaged over the Niño3.4 region and the WWV anomaly averaged over the eastern Pacific. The discharge of warm water from the western Pacific towards the eastern Pacific leads to the initiation of El Niño with a lead period of 6-10 months, and WWV anomaly averaged over the eastern Pacific leads Niño3.4 SST anomaly by 1-2 months. WWV anomalies averaged over PO and WP show above normal values and an increasing tendency during PE (Fig. 9c and k). This increase in anomalous WWV is due to the build-up of heat content in the western Pacific due to the weak westerly or easterly winds during boreal summer and fall months (Fig. 7c) (Allan and D'Arrigo 1999;Levine and McPhaden 2016;Arora and Kumar 2019). Also, anomalous WWV averaged over EP remains positive, indicating persistent El Niño-type conditions throughout the PE duration (1990)(1991)(1992)(1993)(1994) consistent with the previous study (Arora and Kumar 2019) for both observations and CTRL (Fig. 9g). Having validated the CTRL runs, the experimental model runs are now analyzed in detail in the next section.

Changes in surface and subsurface parameters
during model experiments Figure 10 shows the ocean temperature anomaly averaged over the Niño3.4 region for different upper ocean layers for CTRL and eight experimental runs overlaid by observed temperature anomaly. The correlation of ocean temperature anomaly averaged over the Niño3.4 region averaged over layers of upper ocean (5-100 m, 100-200 m, and 200-300 m) between CTRL and observations is 0.81, 0.61, and 0.19, respectively. These correlation values are statistically significant at 99%. A significant correlation between temperature anomaly simulated by OGCM averaged over these layers with observed temperature anomaly implies that the model can capture temperature in the upper ocean over the Niño3.4 region well. Temperature anomalies in the near-surface layer (5-100 m) in the upper ocean are better captured than in the upper ocean's deeper layer (200-300 m). Forcing of normal ocean state (NS:1958(NS: -1962 with horizontal wind forcing at 10 m during protracted ENSO (PE:1990exp5) years and canonical (CE:1980(CE: -1984exp6) ENSO years perturbs mean anomalous state to have variability of PE and CE state respectively. A similar change in variability of PE (CE or CE1) state compared to CTRL is observed when forced with horizontal winds of CE (PE) and NS, respectively. The SST anomaly averaged over the Niño3.4 region during exp7 and exp8 (or exp4 and exp3) shows a shift in variability from canonical El Niño to neutral and protracted ENSO variability, respectively. Regardless of the base state chosen, the results of the model experiment exp7 and exp 8 are consistent with exp4 and exp3, respectively. The effect of the same horizontal wind forcing to different base states (for example, exp1 and exp6) yields similar variability of SST anomaly averaged over the Niño3.4 region with a small difference in magnitude. This difference in magnitude is due to the difference in the mean of base state chosen for the model experiment ( Fig. 2e-h). Table 1 provides a list of experiments performed using OGCM. This interplay of horizontal wind forcing among the ocean's NS, PE, and CE states shows the importance of strength and variability of winds as a governing factor in the onset and type of ENSO. Subsurface layers (100-200 m, 200-300 m) in the upper ocean over the Niño3.4 region also vary following surface layers due to shallow thermocline depth over the equatorial eastern Pacific Ocean (Zelle et al. 2004). Figure 11 shows the ocean temperature anomaly averaged over 5° S-5° N up to 300 m depth for PE, CE, NS, and CE1 for ocean model experiments and observations. The observed PE event shows the strengthening of positive  (Fig. 11a). Compared to PE, the extent of positive anomalies is confined to the eastern Pacific only for CE and CE1 ( Fig. 11b and d). However, the negative subsurface anomalies indicating upwelling Kelvin waves are more zonally extended in CE and CE1 than PE. NS does not show any significant surface or subsurface temperature anomalies (Fig. 11c). The CTRL overestimates the ocean temperature anomalies during NS and PE compared to observations ( Fig. 11e and g). Though the zonal extent of anomalies from the eastern Pacific in the CTRL matches well with observations, subsurface anomalies from the western Pacific are far more zonally extended and strengthened in CTRL than observations, especially in the case of PE and NS. The change of zonal wind forcing over the pre-PE state to the wind forcing used during the CE period (exp1) reduces the magnitude of subsurface temperature anomalies (Fig. 11i). Also, the zonal extent of positive anomalies from the eastern Pacific decreases compared to CTRL and observations. Forcing the pre-PE state with wind forcing during NS (exp2) diminishes the ENSO-like conditions in the equatorial Pacific (Fig. 11m). Similarly, forcing the pre-CE state with wind forcing during PE (NS) shows the signature of PE (NS) in the exp3 (exp4) model experiment ( Fig. 11j and  n). Forcing of pre-NS state with wind forcing of PE (exp5) and CE (exp6) shows weakened PE-and CE-like structure ( Fig. 11k and o) due to the presence of La Niña-like conditions during NS in the CTRL (Fig. 11g). A change of base state from pre-CE to pre-CE1 in exp7 and exp8 also produces similar subsurface thermal evolution as seen in exp4 and exp3, respectively. Based on the model experiments associated with the interplay of wind forcing for 1948-2009, the distinct characteristics of the upper ocean associated with different types of ENSO are compared with observed characteristics of recent protracted ENSO (PE1). The difference in the evolution of near-surface wind anomaly at 10m over WEPO (Fig. 7) and SST anomalies over the Niño3.4 region (Fig. 1) during PE and PE1 can lead to differences in subsurface thermal propagation. Figure 12 shows anomalous WWV up to D20 averaged over PO, EP, and WP using EN4-v4.2.1 for PE1. WWV anomalies averaged over PO show positive values in 2014 and 2015, indicating a heat build-up in the equatorial Pacific Ocean and turning negative in the subsequent year (Fig. 12a). WWV anomalies averaged over EP show smaller  (Fig. 12b). Though the value of anomalous WWV over WP at the beginning of 2014 was positive (Fig. 12c), the weak near-surface westerly wind forcing followed by easterly wind bursts in the summer of 2014 over WP (Fig. 7e) led to a very weak El Niño in 2014 (Fig. 1). Persistent strong anomalous westerlies over EWPO and positive anomalous WWV led to a very strong El Niño in 2015 (Fig. 1). The evolution of WWV anomalies over EP during CE and CE1 exhibits rapid growth in response to strong anomalous westerlies over WP before the onset of El Niño (Fig. 9f and h). However, the build-up of anomalous WWV during PE1 differs somewhat from PE and instead appears to be a blend of CE and PE. Figure 13 shows the ocean temperature anomaly averaged over 5° S-5° N up to 300 m for PE1. The subsurface thermal propagation of temperature anomalies during PE1 shows the strengthening of positive anomalies near the surface up to 100 m depth in eastern Pacific extending to western Pacific instead of up to the dateline in PE (Fig. 11a). The negative subsurface anomalies from the western Pacific are more extended towards the eastern Pacific, similar to subsurface thermal propagation seen in CE (Fig. 11b) and CE1 (Fig. 11d). The model experiments where horizontal wind forcing of PE is applied to pre-CE ocean state (exp3) and pre-CE1 ocean state (exp8) (Fig. 11j and p) tend to produce subsurface propagation similar to PE1 except in the eastern Pacific. Also, the effect of CE forcing to pre-PE ocean state yields similar eastward propagation of negative anomalies from western Pacific to eastern Pacific ( Fig. 11i; exp1). However, the westward extending positive anomalies near the surface in exp1 are weak in magnitude compared to PE1. The thermal propagation along the equatorial Pacific Ocean during PE1 shows protracted (canonical) ENSO-like westward (eastward) propagation near the surface (at subsurface). This mixed behavior is due to the abrupt transition of zonal winds averaged over the EWPO from hovering around the mean in 2014 to strong westerlies in 2015 (Fig. 7e). Yang et al. (2018) compared two strong ENSO events in 2015-2016 and 1997-1998 with a comparable magnitude of Niño3.4 index (also shown in Fig. 1) and showed relatively weak zonal SST gradient leading to weak air-sea coupling in compared to 1997-1998consistent with Hu et al. (2020. The weakened discharge of WWV from WP towards EP during the weak El Niño of 2014 keeps the heat content high in the entire equatorial Pacific Ocean (Fig. 12). The magnitude of negative subsurface anomalies from the western Pacific is weakened and is far more extended towards the east during PE1 than PE. Previous studies have also documented the slowing down of the development of warm SST in 2014-2015 due to air-sea feedback between SST anomalies and zonal wind anomalies (Levine and McPhaden 2016;Zhu et al., 2016;Lim et al. 2017;Puy et al. 2017;Hu et al. 2020;Zhao et al. 2021). None of the model experiments

Summary and conclusion
Previous studies have reported six protracted (also termed as persistent) El Niño events (1894-1897, 1911-1914, 1939-1942, 1990-1995, 2002-2006, and 2014-2016) since 1876 (Allan and D'Arrigo 1999;Arora and Kumar 2019;Allan et al. 2020). It is certain from these records that the frequency of occurrence of these protracted events has increased during the last two decades. This study attempts to understand the life cycle of the two primitive types of El Niño (canonical and protracted) using an ocean model. Anomalous westerlies and heat content over the WP and Maritime continent lead to the onset of El Niño in the eastern Pacific. The magnitude and frequency of these anomalous westerlies determine the type of ENSO state and associated teleconnections (Arora and Kumar 2019). Three different states termed CE, PE, and NS spanning over 5 years are identified using the standard definition of El Niño by CPC-NOAA. Each type of state exhibits a peculiar surface and subsurface temperature structure in the tropics before the onset of ENSO. The two extreme canonical El Niño events (CE and CE1) are similar in mean spatial features and associated air-sea interaction mechanisms in the study period. CTRL simulates higher SST values over WP and more zonally extended cold tongue in the EP before the onset of the PE compared to CE and NS state consistent with previous studies (Levine and McPhaden 2016). The ocean model used in this study exhibits systematic bias at the surface and subsurface, yielding an increased east-west gradient of SST along the equator in the Pacific Ocean compared to observations. There is a difference in the distribution of anomalous surface forcing over WP during different states. While PE state follows a normal distribution, NS and CE state show skewness towards positive values indicating more anomalous westerlies during NS and CE. Model experiments are performed by interchanging the surface wind forcing among different ENSO states without altering the ocean state before the onset of El Niño. While chosen initial ocean state keeps the memory of the mean state, the selection of the applied surface wind forcing chosen among the three types of forcings determines the variability of the upper ocean. The wind forcing at the surface in the boreal summer months can determine the fate of ongoing subsurface El Niño and associated teleconnections.
Though the onset of PE1 began with canonical-like increasing anomalous WWV and westerlies over equatorial WP, the absence of strong westerlies at low latitudes in the summer months of 2014 led to a weak borderline El Niño leading to a reduction of discharge of anomalous WWV from WP towards EP. The earlier presence of anomalous equatorial Pacific heating and intense downwelling Kelvin wave due to strong anomalous westerlies in 2015 led to a strong El Niño with a comparable magnitude of peak SST anomalies over Niño3.4 during CE and CE1. The impact of the horizontal wind forcing during PE to ocean state before the onset of canonical El Niño yields subsurface propagation similar to PE1 (exp3 and exp8). The impact of CE forcing to ocean state before the onset of PE yields similar (weakened) subsurface propagation of negative anomalies from WP (EP) to EP (WP) (exp1). The evolution of SST anomalies over the equatorial eastern Pacific and subsurface propagation of downwelling Rossby and upwelling Kelvin waves during PE1 is a hybrid of PE and CE states. An improved understanding of underlying processes leading to the biases in the upper ocean in climate models can help understand the processes involved during the evolution and decay phase of El Niño.